Set up and cut over custom domains for API Gateway
npm install serverless-plugin-custom-domainThis is a plugin for Serverless that injects a CloudFormation Custom Resource in your deployed stack that sets up a base path mapping between the ApiGateway::Deployment that Serverless creates, and an API Gateway Custom Domain.
This version uses a newer version of add-custom-resource which requires a newer Node.js runtime.
This release brings in a new version of add-custom-resource that updates the runtime version of Lambda. This means your custom resources may re-execute, and should be idempotent. No actual changes in this library have been made, but this is just to be cautious.
A bug in 1.x caused an extra, unused LogGroup to be created, called /aws/lambdaservice-stage-CustomBasePathMa-- instead of /aws/lambda/service-stage-CustomBasePathMa--. However, AWS automatically creates a LogGroup for your functions with the correct name. This means that any deployments using 1.x have two LogGroups. After upgrading to 2.x+, deployments may fail due to a CloudFormation error that "the log group already exists". Sorry about this! It doesn't always happen, but thankfully, the fix is pretty straightforward if it does and it should only happen the one time.
- delete the log group (it's the 'correctly' named one, e.g. /aws/lambda/service-stage-CustomBasePathMa--)
- redeploy!
- feel free to delete the other, incorrectly named log group, it was never used too
``yaml
service: my-service
plugins:
- serverless-plugin-custom-domain
custom:
domain: "${opt:region}.myservice.foo.com"
`
custom.domain can also be an object, with the following propeties:
- name: the domain name, same as the above string e.g. ${opt:region}.myservice.foo.com
- a custom base path, instead of the default (none) - a base path is a prefix e.g. /v1
CloudFormation supports ApiGateway::BasePathMapping resources but I found they frequently fail to update correctly. Implementing the (relatively simple) logic to get-and-update-or-create combined with a remove` hook for cleanup has proven to be more reliable.
These take a long time to provision and are long-lived persistent resources that have Route53 entires pointing at them as well as ACM certificates that have to be requested and approved. You should manage these outside of Serverless, either via CloudFormation or something like Terraform.